Return-Path: X-Spam-Checker-Version: SpamAssassin 3.4.0 (2014-02-07) on aws-us-west-2-korg-lkml-1.web.codeaurora.org Received: from lists.sourceforge.net (lists.sourceforge.net [216.105.38.7]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by smtp.lore.kernel.org (Postfix) with ESMTPS id 002E2C54F57 for ; Tue, 28 Jul 2026 11:49:17 +0000 (UTC) DKIM-Signature: v=1; a=rsa-sha256; q=dns/txt; c=relaxed/relaxed; d=lists.sourceforge.net; s=beta; h=Content-Transfer-Encoding:Content-Type:Cc: List-Subscribe:List-Help:List-Post:List-Archive:List-Unsubscribe:List-Id: Subject:MIME-Version:References:In-Reply-To:Message-ID:Date:To:From:Sender: Reply-To:Content-ID:Content-Description:Resent-Date:Resent-From:Resent-Sender :Resent-To:Resent-Cc:Resent-Message-ID:List-Owner; bh=BmObCGVUYDmiGhAK/avuIXvOiABjKr8EvGr7fTGKyts=; b=JOTsJkW4Q0b4IyuyrPyWxluQaI K+JVWIkXh4MfJCav5jteQUvn2hDme4zT1DBKN76go+5JalqwpxTCoRDLTMOaIljENvOroLunQ39b/ aDCqZevRbjcucYYZ4KMgh209pZ0PTJduN1r6VojUIY+QJyeAcLSmIkuUK8scpp5iQoF4=; Received: from [127.0.0.1] (helo=sfs-ml-4.v29.lw.sourceforge.com) by sfs-ml-4.v29.lw.sourceforge.com with esmtp (Exim 4.95) (envelope-from ) id 1wogJA-0002os-HD; Tue, 28 Jul 2026 11:49:16 +0000 Received: from [172.30.29.66] (helo=mx.sourceforge.net) by sfs-ml-4.v29.lw.sourceforge.com with esmtps (TLS1.2) tls TLS_ECDHE_RSA_WITH_AES_256_GCM_SHA384 (Exim 4.95) (envelope-from ) id 1wogJ9-0002oe-CW for openvpn-devel@lists.sourceforge.net; Tue, 28 Jul 2026 11:49:15 +0000 DKIM-Signature: v=1; a=rsa-sha256; q=dns/txt; c=relaxed/relaxed; d=sourceforge.net; s=x; h=Content-Transfer-Encoding:MIME-Version:References: In-Reply-To:Message-ID:Date:Subject:Cc:To:From:Sender:Reply-To:Content-Type: Content-ID:Content-Description:Resent-Date:Resent-From:Resent-Sender: Resent-To:Resent-Cc:Resent-Message-ID:List-Id:List-Help:List-Unsubscribe: List-Subscribe:List-Post:List-Owner:List-Archive; bh=Yg5CoNDMDrOG6jpu826kHHRn6bPkRuc1+5tmPj7qUS4=; b=LRBa0vTK1le5a3xXoSpElqSzV0 DQgN9HyxgJ0smZhr9IelKJuecLdinXdQj4eUNo3CJsOvr8fyzawWvcSlfazI4czast8P52bZUYRna zr9icp4RDB8DJbeWdQnDZUNjlGbtJN14TXqVaTCJ390YRng7yiNCzEYmezVxAvmnbXKE=; DKIM-Signature: v=1; a=rsa-sha256; q=dns/txt; c=relaxed/relaxed; d=sf.net; s=x ; h=Content-Transfer-Encoding:MIME-Version:References:In-Reply-To:Message-ID: Date:Subject:Cc:To:From:Sender:Reply-To:Content-Type:Content-ID: Content-Description:Resent-Date:Resent-From:Resent-Sender:Resent-To:Resent-Cc :Resent-Message-ID:List-Id:List-Help:List-Unsubscribe:List-Subscribe: List-Post:List-Owner:List-Archive; bh=Yg5CoNDMDrOG6jpu826kHHRn6bPkRuc1+5tmPj7qUS4=; b=jG1EkRP08i62VVf0KwbsJwst7y fKmBV6pLJxWH6ABtudL9cFOgn1m3jQWO/PtkGJK3F+RPuI2BKhWHksy4Mn6sIYMvhRnRs1KrCFsu4 Z1fk5EqZdKvNoXK+e6tO/KayeE1/mj3l8YIv61OZwqk/88HBIlIlApGDeuSUediR04XM=; Received: from mout-p-101.mailbox.org ([80.241.56.151]) by sfi-mx-1.v28.lw.sourceforge.com with esmtps (TLS1.2:ECDHE-RSA-AES256-GCM-SHA384:256) (Exim 4.95) id 1wogJB-0003C7-SY for openvpn-devel@lists.sourceforge.net; Tue, 28 Jul 2026 11:49:15 +0000 Received: from smtp1.mailbox.org (unknown [10.196.197.1]) (using TLSv1.3 with cipher TLS_AES_256_GCM_SHA384 (256/256 bits) key-exchange x25519 server-signature RSA-PSS (4096 bits) server-digest SHA256) (No client certificate requested) by mout-p-101.mailbox.org (Postfix) with ESMTPS id 4h8Ydt3lCgz8v0r; Tue, 28 Jul 2026 13:49:06 +0200 (CEST) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=unstable.cc; s=MBO0001; t=1785239346; h=from:from:reply-to:subject:subject:date:date:message-id:message-id: to:to:cc:cc:mime-version:mime-version: content-transfer-encoding:content-transfer-encoding: in-reply-to:in-reply-to:references:references; bh=Yg5CoNDMDrOG6jpu826kHHRn6bPkRuc1+5tmPj7qUS4=; b=JiN4gEt04iu5R5Pb+me7Qj/olFlLcXhRGF8xxmFSauLdM8xiXigKQ1etNAV0wzZstVtYBA CH1pzvXiXDm4RUWYpEL227DIoDuN9ZnnAHmSHH3A0X027qNvlTtzUXspggBPI2PIhBj065 zb7AD017dDSEFpLBRGVIK8/LVNt2PCwG7NHliUDQ0FqldKtH+fweEVp8hSziF5N5KEF3Bi v9WSMjHcXksHyoS/Q3GFqlP0z5bfjpSjtqnxzqD/URZxsJ7NIchFdBtBcZ89Sz2rc8q2IU 0ETQCVCdZR8zHAxbAZEZr2QLaA7unQesmKskxLw6fJYEt3CJRF+Gi7ohMKN/MQ== From: Antonio Quartulli To: openvpn-devel@lists.sourceforge.net Date: Tue, 28 Jul 2026 13:48:55 +0200 Message-ID: <20260728114855.1323861-10-a@unstable.cc> In-Reply-To: <20260728114855.1323861-1-a@unstable.cc> References: <20260728114855.1323861-1-a@unstable.cc> MIME-Version: 1.0 X-Headers-End: 1wogJB-0003C7-SY Subject: [Openvpn-devel] [PATCH ovpn net v4 9/9] ovpn: invalidate the UDP TX dst_cache when the flow key changes X-BeenThere: openvpn-devel@lists.sourceforge.net X-Mailman-Version: 2.1.21 Precedence: list List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Cc: Antonio Quartulli Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit Errors-To: openvpn-devel-bounces@lists.sourceforge.net From: Antonio Quartulli ovpn_udp{4,6}_output() resolve a route from a flow key sampled from the peer binding and the transport socket, then cache the result in the per-peer dst_cache. Several of those sources may change concurrently with TX, and the dst_cache epoch (reset_ts vs the per-CPU refresh_ts stamped at get-miss time) only neutralizes the common ordering. Three issues remain: - ovpn_peer_endpoints_update() may either update bind->local in place or replace the whole bind via RCU (float -> new remote, hence new daddr/dport/oif). It already dst_cache_reset()s, but the TX path can still cache a dst it resolved with the pre-update values if its dst_cache_get-miss lands a strictly later jiffy than the reset. - inet_sk(sk)->inet_sport can be reset to 0 by __udp_disconnect() (connect() with AF_UNSPEC) on a socket without SOCK_BINDPORT_LOCK, and sk->sk_mark can change any time via setsockopt(SO_MARK). Neither triggers an ovpn cache reset, so a previously-cached entry resolved with the old value persists until dst obsolescence. Both fields are also read locklessly into the flow key (data race). - A sport of 0 means the transport socket has been disconnected and unhashed; sending a UDP packet from source port 0 is nonsense. In the common dispatcher ovpn_udp_output() (so every TX, including cache hits, runs the check): - Sample inet_sport with READ_ONCE(). If it is 0, emit a one-time netdev_warn_once() and return -EIO so ovpn_udp_send_skb() drops the skb. - Sample sk_mark with READ_ONCE(). - Compare both against the values stored when the dst_cache was last (re-)populated (new per-peer fields dst_cache_sport/dst_cache_mark, zero-initialised by kzalloc_obj() in ovpn_peer_new()). On mismatch dst_cache_reset() the cache and WRITE_ONCE() the new values, so the subsequent dst_cache_get() misses and the lookup re-resolves with the current sport/mark. - Pass sport/mark down to ovpn_udp{4,6}_output(); they use those in the flowi initializer and skip the per-function sampling. The post-lookup re-check in the v4/v6 paths covers both the bind/local race the original commit addressed (rcu_access_pointer on peer->bind and READ_ONCE/ovpn_peer_local_ipv6 on bind->local) and sport/mark: the TX-entry check alone is not enough, because a slow resolver can finish its route lookup after another CPU has already re-tagged the cache with a different sport/mark, so it must re-verify both before populating the cache. sk_protocol is immutable post-creation and is intentionally read plain. The in-flight packet is still transmitted with the resolved parameters; only the cache is guarded. No fast-path lock is added. Fixes: 08857b5ec5d9 ("ovpn: implement basic TX path (UDP)") Signed-off-by: Antonio Quartulli --- drivers/net/ovpn/peer.h | 8 ++++ drivers/net/ovpn/udp.c | 101 ++++++++++++++++++++++++++++++++++------ 2 files changed, 94 insertions(+), 15 deletions(-) diff --git a/drivers/net/ovpn/peer.h b/drivers/net/ovpn/peer.h index c0994c606554..17d57b12fa5e 100644 --- a/drivers/net/ovpn/peer.h +++ b/drivers/net/ovpn/peer.h @@ -46,6 +46,12 @@ * @tcp.sk_cb.ops: pointer to the original prot_ops object (TCP only) * @crypto: the crypto configuration (ciphers, keys, etc..) * @dst_cache: cache for dst_entry used to send to peer + * @dst_cache_sport: inet_sport observed when the dst_cache was last + * (re-)populated; compared on every TX to detect changes + * (e.g. connect(AF_UNSPEC)) and invalidate the cache + * @dst_cache_mark: sk_mark observed when the dst_cache was last + * (re-)populated; compared on every TX to detect changes + * via setsockopt(SO_MARK) and invalidate the cache * @bind: remote peer binding * @keepalive_interval: seconds after which a new keepalive should be sent * @keepalive_xmit_exp: future timestamp when next keepalive should be sent @@ -102,6 +108,8 @@ struct ovpn_peer { } tcp; struct ovpn_crypto_state crypto; struct dst_cache dst_cache; + __be16 dst_cache_sport; + u32 dst_cache_mark; struct ovpn_bind __rcu *bind; unsigned long keepalive_interval; unsigned long keepalive_xmit_exp; diff --git a/drivers/net/ovpn/udp.c b/drivers/net/ovpn/udp.c index 17d65d1595ed..fc5ef77d17b0 100644 --- a/drivers/net/ovpn/udp.c +++ b/drivers/net/ovpn/udp.c @@ -143,19 +143,24 @@ static int ovpn_udp_encap_recv(struct sock *sk, struct sk_buff *skb) */ static int ovpn_udp4_output(struct ovpn_peer *peer, struct ovpn_bind *bind, struct dst_cache *cache, struct sock *sk, - struct sk_buff *skb) + struct sk_buff *skb, __be16 sport, u32 mark) { + /* bind->local is updated in place under peer->lock; a single aligned + * word is read/written atomically via {READ,WRITE}_ONCE. Snapshot it + * so the post-lookup validity check can compare against the value we + * actually resolved with: fl.saddr may be overwritten by the FIB when + * the local address is unset, and comparing bind->local against that + * would spuriously skip caching. + */ + __be32 saddr = READ_ONCE(bind->local.ipv4.s_addr); struct rtable *rt; struct flowi4 fl = { - /* bind->local is updated in place under peer->lock; a single - * aligned word is read/written atomically via {READ,WRITE}_ONCE - */ - .saddr = READ_ONCE(bind->local.ipv4.s_addr), + .saddr = saddr, .daddr = bind->remote.in4.sin_addr.s_addr, - .fl4_sport = inet_sk(sk)->inet_sport, + .fl4_sport = sport, .fl4_dport = bind->remote.in4.sin_port, .flowi4_proto = sk->sk_protocol, - .flowi4_mark = sk->sk_mark, + .flowi4_mark = mark, }; int ret; @@ -171,6 +176,7 @@ static int ovpn_udp4_output(struct ovpn_peer *peer, struct ovpn_bind *bind, * new look up */ fl.saddr = 0; + saddr = 0; spin_lock_bh(&peer->lock); WRITE_ONCE(bind->local.ipv4.s_addr, 0); spin_unlock_bh(&peer->lock); @@ -180,6 +186,7 @@ static int ovpn_udp4_output(struct ovpn_peer *peer, struct ovpn_bind *bind, rt = ip_route_output_flow(sock_net(sk), &fl, sk); if (IS_ERR(rt) && PTR_ERR(rt) == -EINVAL) { fl.saddr = 0; + saddr = 0; spin_lock_bh(&peer->lock); WRITE_ONCE(bind->local.ipv4.s_addr, 0); spin_unlock_bh(&peer->lock); @@ -196,7 +203,21 @@ static int ovpn_udp4_output(struct ovpn_peer *peer, struct ovpn_bind *bind, ret); goto err; } - dst_cache_set_ip4(cache, &rt->dst, fl.saddr); + /* only cache the result if the parameters we resolved with are still + * current: a concurrent ovpn_peer_endpoints_update() may have replaced + * the bind (float) or updated bind->local in place, and a concurrent + * ovpn_udp_output() may have re-tagged the cache with a different + * sport/mark after we sampled them. In any of these cases the cache was + * already reset and re-caching a route resolved with the stale values + * would poison it, so bail out and reset instead. + */ + if (rcu_access_pointer(peer->bind) == bind && + READ_ONCE(bind->local.ipv4.s_addr) == saddr && + READ_ONCE(peer->dst_cache_sport) == sport && + READ_ONCE(peer->dst_cache_mark) == mark) + dst_cache_set_ip4(cache, &rt->dst, fl.saddr); + else + dst_cache_reset(cache); transmit: udp_tunnel_xmit_skb(rt, sk, skb, fl.saddr, fl.daddr, 0, @@ -221,24 +242,30 @@ static int ovpn_udp4_output(struct ovpn_peer *peer, struct ovpn_bind *bind, */ static int ovpn_udp6_output(struct ovpn_peer *peer, struct ovpn_bind *bind, struct dst_cache *cache, struct sock *sk, - struct sk_buff *skb) + struct sk_buff *skb, __be16 sport, u32 mark) { struct dst_entry *dst; + struct in6_addr local, saddr; int ret; struct flowi6 fl = { .daddr = bind->remote.in6.sin6_addr, - .fl6_sport = inet_sk(sk)->inet_sport, + .fl6_sport = sport, .fl6_dport = bind->remote.in6.sin6_port, .flowi6_proto = sk->sk_protocol, - .flowi6_mark = sk->sk_mark, + .flowi6_mark = mark, .flowi6_oif = bind->remote.in6.sin6_scope_id, }; /* bind->local is updated in place under peer->lock; read the 128-bit - * address under the peer seqcount to avoid a torn read + * address under the peer seqcount to avoid a torn read. Snapshot it so + * the post-lookup validity check can compare against the value we + * actually resolved with: fl.saddr may be overwritten by the FIB when + * the local address is unset, and comparing bind->local against that + * would spuriously skip caching. */ ovpn_peer_local_ipv6(peer, bind, &fl.saddr); + saddr = fl.saddr; local_bh_disable(); dst = dst_cache_get_ip6(cache, &fl.saddr); @@ -251,6 +278,7 @@ static int ovpn_udp6_output(struct ovpn_peer *peer, struct ovpn_bind *bind, * new look up */ fl.saddr = in6addr_any; + saddr = in6addr_any; spin_lock_bh(&peer->lock); write_seqcount_begin(&peer->bind_local_seq); bind->local.ipv6 = in6addr_any; @@ -267,7 +295,22 @@ static int ovpn_udp6_output(struct ovpn_peer *peer, struct ovpn_bind *bind, &bind->remote.in6, ret); goto err; } - dst_cache_set_ip6(cache, dst, &fl.saddr); + /* only cache the result if the parameters we resolved with are still + * current: a concurrent ovpn_peer_endpoints_update() may have replaced + * the bind (float) or updated bind->local in place, and a concurrent + * ovpn_udp_output() may have re-tagged the cache with a different + * sport/mark after we sampled them. In any of these cases the cache was + * already reset and re-caching a route resolved with the stale values + * would poison it, so bail out and reset instead. + */ + ovpn_peer_local_ipv6(peer, bind, &local); + if (rcu_access_pointer(peer->bind) == bind && + ipv6_addr_equal(&local, &saddr) && + READ_ONCE(peer->dst_cache_sport) == sport && + READ_ONCE(peer->dst_cache_mark) == mark) + dst_cache_set_ip6(cache, dst, &fl.saddr); + else + dst_cache_reset(cache); transmit: /* user IPv6 packets may be larger than the transport interface @@ -306,12 +349,40 @@ static int ovpn_udp_output(struct ovpn_peer *peer, struct dst_cache *cache, struct sock *sk, struct sk_buff *skb) { struct ovpn_bind *bind; + __be16 sport; + u32 mark; int ret; /* set sk to null if skb is already orphaned */ if (!skb->destructor) skb->sk = NULL; + sport = READ_ONCE(inet_sk(sk)->inet_sport); + if (unlikely(!sport)) { + /* the transport UDP socket has been disconnected (e.g. via + * connect(AF_UNSPEC)): inet_sport == 0 means the socket has + * been unhashed and sending from source port 0 is nonsense; + * refuse and tell the operator + */ + netdev_warn_once(peer->ovpn->dev, + "UDP transport socket has no source port; was it disconnected?\n"); + return -EIO; + } + mark = READ_ONCE(sk->sk_mark); + + /* userspace can change sk_mark (via setsockopt(SO_MARK)) and + * inet_sport (via connect(AF_UNSPEC)) at any time without notifying + * ovpn; if either differs from what the dst_cache was last populated + * with, invalidate the cache now so a hit doesn't return a dst + * resolved with the old value + */ + if (READ_ONCE(peer->dst_cache_sport) != sport || + READ_ONCE(peer->dst_cache_mark) != mark) { + dst_cache_reset(cache); + WRITE_ONCE(peer->dst_cache_sport, sport); + WRITE_ONCE(peer->dst_cache_mark, mark); + } + rcu_read_lock(); bind = rcu_dereference(peer->bind); if (unlikely(!bind)) { @@ -323,11 +394,11 @@ static int ovpn_udp_output(struct ovpn_peer *peer, struct dst_cache *cache, switch (bind->remote.in4.sin_family) { case AF_INET: - ret = ovpn_udp4_output(peer, bind, cache, sk, skb); + ret = ovpn_udp4_output(peer, bind, cache, sk, skb, sport, mark); break; #if IS_ENABLED(CONFIG_IPV6) case AF_INET6: - ret = ovpn_udp6_output(peer, bind, cache, sk, skb); + ret = ovpn_udp6_output(peer, bind, cache, sk, skb, sport, mark); break; #endif default: -- 2.54.0 _______________________________________________ Openvpn-devel mailing list Openvpn-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/openvpn-devel